Now that we have best practices, let's automate them: Chodera's Yank. ⢠Automates alchemical free energy calculations in implicit or explicit solvent. ⢠Hydration ...
Automating best practices: Good for users, good for science
David Mobley
Hydration free energies are a case where we have good established practices we automate
Hydration free energy predictions have improved greatly over the years
Hydration free energy predictions have improved greatly over the years
Hydration free energy predictions have improved greatly over the years
Now that we have best practices, let’s automate them: Chodera’s Yank
• • •
Automates alchemical free energy calculations in implicit or explicit solvent Hydration or binding We put it on Orion: We have a hydration floe and a draft binding floe
And we have our FreeSolv database of calculated and experimental hydration free energies
•
• •
643 curated hydration free energies for neutral compounds, with literature references, structures, etc. github.com/mobleylab/freesolv We put it on Orion
So, now we have a Floe for hydration free energies and it will take FreeSolv as input
•
We ran in implicit (GBSA) solvent as an initial test
Explicit solvent is working also
Explicit solvent is working also
Outcome: Something that took 200K+ CPU hours over a month, plus a couple of weeks of human time now takes a button press and a few hours
The same technology allows for easy automation of relative solubility prediction A
Solvent C
Solvent B A
A A
A A A A
A A A A
A A A A
A
A A A A
A A A A
A A A A
A A A A
A A A A
Relative solubilities can be calculated fairly accurately in arbitrary organic solvents
• •
Process chemistry (separations) Force field testing
Transfer
Partition and distribution coefficients are also accessible via solvation calculations
−ΔGTransfer log P = RT ln10 ΔGTransfer = ΔGSolvation − ΔGHydration
• • •
Passive membrane permeability Separations Force field testing
Transfer
Partition and distribution coefficients are also accessible via solvation calculations
−ΔGTransfer log P = RT ln10 ΔGTransfer = ΔGSolvation − ΔGHydration
• • •
Passive membrane permeability Separations Force field testing
SAMPL5 focused on calculating cyclohexanewater logD, which we could do on Orion Some of the compounds considered
We’ve had a dire need for good automated workflows with interchangeable components; containers and standard data formats Orion is enabling that allow us to evaluate the technology standardized! data formats
standardized! data formats
industry! datasets
standard
benchmarks preparation! pipeline
modeling tool
automated! analysis/! evaluation
It’s interchanging the components which will reallypractices advance our science best can be
evaluated by testing variations standardized! data formats
standardized! data formats
industry! datasets
standard
benchmarks
modeling tool preparation! pipeline! variations
automated! analysis/! evaluation
We need to be able to use the latest science in existing workflows to see how much it helps, rather than waiting years (if ever) for software company X to decide it’s valuable and incorporate it in a polished application Orion can change that — we can drop a new method into an existing floe and test immediately Side benefit: If your speciality is in the area of one cube, you don’t have to be responsible for the rest of the floe
One thing to (try to) avoid: forming new silos around many incompatible workflow engines so we STILL can’t compare the latest methods head-tohead easily •
Old silos: simulation packages/modeling ecosystems
AMBER
CHARMM
CHARMM
Desmond
….
One thing to (try to) avoid: forming new silos around many incompatible workflow engines so we STILL can’t compare the latest methods head-tohead easily •
Old silos: simulation packages/modeling ecosystems
AMBER •
CHARMM
CHARMM
Desmond
New silos: workflows/pipelines
Orion
MDT (?)
…
….
….
Orion is pretty exciting
Orion is pretty exciting
Automation is great
Orion is pretty exciting
Automation is great
Large scale hydration for anyone!
Orion is pretty exciting
Automation is great
Large scale hydration for anyone!
SAMPL5_050
OH N
N N
SAMPL5_050_conformer2
O N
N
N N
N H
We can calculate distribution, partition, and relative solubility
Orion is pretty exciting
Automation is great
Large scale hydration for anyone!
best practices can be evaluated by testing variations
SAMPL5_050
OH N
N N
standardized! data formats
SAMPL5_050_conformer2
standardized data formats
O N
N
standard
benchmarks
N N modeling tool
N H
We can calculate distribution, partition, and relative solubility
preparation! pipeline! variations
Automation isn’t just for users, but enables science
Acknowledgments • Distribution coefficients: JW Feng, Baiwei Lin, Dan Ortwine, Joe Pease, Justin Dancer, Bas Rustenberg, John Chodera • Yank: John Chodera, Andrea Rizzi, Levi Naden, Chodera lab • Mobley lab present and former including highlighted work from Nathan Lim, Guilherme Matos, Caitlin Bannan, Kalli Burley, Shuai Liu • OpenEye Orion crew (esp. Craig, Forrest, Matt) & Ant • Christopher Bayly • Many, many collaborators including Chris Fennell, Michael Shirts, Mike Gilson, Andrew Paluch, J. Peter Guthrie, and many others.
We also looked at excess solubility in binary mixtures of water/ethanol
We also looked at excess solubility in binary mixtures of water/ethanol
